Stage 1: input and research
You provide a topic, an outline or a finished script. For topics, a language model with web search can gather sources and draft an outline; the sources should be shown to you so claims can be checked.
Stage 2: script and storyboard
The script is written in short spoken sentences, then grouped into scenes of roughly four to ten seconds. Each scene records its narration, a description of the picture, and a camera move such as a slow zoom.
Stage 3: pictures
Each picture description becomes an image in your chosen style. A character reference image can be reused so the same person or mascot appears in several scenes, although perfect consistency is not guaranteed and should be reviewed.
Stage 4: voice and timing
Narration is spoken sentence by sentence. Because each sentence's audio length is known exactly, scene durations and caption timing follow from it without guessing.
Stage 5: rendering
A renderer places each image for its sentence duration, applies the camera move, lays the voice and optional music underneath, burns in or exports captions, and writes the MP4.
Where it goes wrong
- Facts in a generated script that nobody checked
- Characters that look different between scenes
- Text inside images, which image models often misspell
- Voices that mispronounce names